madVR - high quality video renderer (GPU assisted)

[madshi]

Hm, I don't see any options for it there - only "debanding - toggle" and "debanding custom settings - toggle".
But by mashing the keyboard I figured out that new shortcuts are ctrl+shift+arrows on numpad :)
Ah yes, you're right arrows for switching, but the shortcut to activate the whole deband menu can be configured.

i mean 4:2:0 -> 4:4:4.

isn't SuperRes a lot slower?
For chroma it's pretty fast. Probably faster than NEDI would be.

Matching framerate seems to be fine (60 @ 60, anyway), I produced the problem playing 24p on a 60 Hz screen.
Ah, that's what I thought. That's sad, though, it means that one part of the DXGI presentation API doesn't work properly on some PCs. <sigh>

Man, you really need a good European TV with 23.976, 24, 25, 29.97, 30, 50, 59.94 and 60 Hz.
Well, I have a multi-refresh rate projector which does all that, but it's in my cellar, and hard to develop on.

Yes. MPC-HC's window needs to be fully covered by another window or minimized, then it freezes after ~30 seconds being like that. Sometimes it doesn't appear frozen after getting back to\unminimizing it and instead very quickly shows all the frames it supposed to show while being stuck, OSD reports 900+ dropped frames and the playback continues. Pretty weird. I PM'd you a freeze report.
Freeze reported doesn't say anything. FWIW, I can let MPC-HC sit minimized with D3D11 on my PC without any problems. Zero frame drops. Which OS were you on again? If win7: Do you have Aero disabled or enabled?

When the next build is released, can you upload a debug log for me with that, just with frame drops caused by running in minimized state, doesn't have to be the freeze. Doesn't also have to be 30 seconds. Just long enough so that I can see that there's any kind of trouble in minimized state...

Yes.
madVR high:

Shiandow 0,7; 2 (MPDN default):

However, Shiandow's deband can be very interesting with low quality sources, since it can be used as some kind of deblocking filter.
Ah, thanks. Hmmmm... Deblocking? I think it's better to create a dedicated deblocking filter for that. Though, some of the ideas of Shiandow's deband algorithm could be useful for deblocking, too.

[Werewolfy]

Thanks for the feedback. Ok, so the "new" algorithm can't compete with the old "low" algo, if you aim for max detail preservation. But how does the new algo compare to the old "high" algo? From what Shiandow has written so far I think that's the likeliest situation where the new algo could complete (and maybe beat) with the old one.

I can't really find a situation where the new algo beat the old high algo. In some areas of the picture it's a little bit better but on the others areas it's worse.

I used threshold : 0.9 and detail level : 3 to compare and even tested with 1.0 and 2 to see if it makes a difference but I always prefer the high algo. The details are much more well preserved and I don't know why but the high algo just feels more natural to my eyes. With the new one, sometimes I see banding turned into some sort of "clouds".


I hear you. What you're asking for is not so easy to implement, though. For most users (I'd say 99% or more) it's much better to have a "low/mid/high" switch for every important algorithm then to have a dozen different knobs and dials. Most users wouldn't know what to do with so many options and would be overwhelmed. I understand that some advanced users like you would like to have more control. But let's be honest here:

How many of you are using the "debanding custom settings - toggle" keyboard shortcut which madVR currently offers, in order to fine tune those 5 or 6 different debanding parameters? Seriously, does *anybody* use it? I doubt it's more than 5 out of all madVR users.

One thing I could imagine is adding an advanced settings page somewhere where all the hundreds of different options can be fine tuned. However, it's not really easy, because the main dialog will continue to have low/mid/high settings. So would the advanced options have to allow you to redefine what low/mid/high means exactly? That would increase the number of options by a factor of 3, because every option would have to be available for low, mid and high presets. <sigh>

Anyway, something like this is not coming any time soon. It's too time consuming. For now I want to find proper simple defaults like low,mid,high and I *will* remove the other options. Sorry. I have to set priorities on the benefits of the majority of users. And the majority wants an easy to understand configuration dialog.


I agree with your thinking. When I tested your debanding filter, I felt that custom settings needed to stay. But to be honest, I never used it after my testings done...

Maybe just a custom preset like you find in videogames in the addition of the low, medium and high preset would be a good idea to satisfy everyone? In that case you wouldn't have a ton of options to deal with and it wouldn't bother the average user. It's just an idea, I don't think I will personnaly use it.

I justed want to add that when 10bit videos and UHD Blu-Ray are available, the debanding filter may have to be fine tuned. But I think we will see that in time :)

[MS-DOS]

Freeze reported doesn't say anything. FWIW, I can let MPC-HC sit minimized with D3D11 on my PC without any problems. Zero frame drops. Which OS were you on again? If win7: Do you have Aero disabled or enabled?

When the next build is released, can you upload a debug log for me with that, just with frame drops caused by running in minimized state, doesn't have to be the freeze. Doesn't also have to be 30 seconds. Just long enough so that I can see that there's any kind of trouble in minimized state...

7, Aero enabled. If I choose "disable desktop composition" in MadVR with D3D11 enabled the image just flickers between the first frame and black, while on D3D9 it falls back to the old path and plays normally.

Will do.

BTW, I had a problem before, that appeared in some of the previous builds and was fixed in some of the later ones (sorry for being that exact) - when the player was minimized and the video finished playing it didn't move to the next file in the playlist and got stuck. It didn't happen with EVR. Guess what, the same now happens with D3D11 activated, but now it immediately unstucks when I unminimize the player. Actually, the audio will unstuck even if I hover the mouse over mpc-hc's taskbar icon so that the preview will appear, but the image there will remain frozen.

[Moony349]

Other people are still getting presentation glitches in FSE mode on Windows 7 using x64 and Dx11, right?

[Siso]

Other people are still getting presentation glitches in FSE mode on Windows 7 using x64 and Dx11, right?

Presentation glitches no, but weird rendering times and present times, also present queues and render queues are not filled...

[noee]

Other people are still getting presentation glitches in FSE mode on Windows 7 using x64 and Dx11, right?

Not here. Though I'm running dual-mon with playback on my second mon, fwiw. 0,0,0.

However, with FireFox and OMTC turned on, there are issues.

[Moony349]

Matching framerate seems to be fine (60 @ 60, anyway), I produced the problem playing 24p on a 60 Hz screen.

Seems like this is the issue I'm having.

[mark0077]

Hi madshi, I've downloaded the latest version of madVR, and thanks for it.

I've been testing D3D11 for presentation. In windowed mode all is fine, but in FSE mode I get seemingly exactly 60 presentation glitches per second. I'm playing back a 24fps movie but converted to 60fps with SVP, on a 60hz screen.

This is using 32bit mpc-be, madVR, LAV Filters, ffdshow + SVP. In windows 8.1 64bit. NVidia GTX 980 with latest WHQL drivers. Any more testing / info needed just let me know.

(Note: I see presentation queue is at 2-4 / 8, incase that gives any more information)

[JarrettH]

Not enough information. Windowed mode or fullscreen windowed mode or FSE mode? All the time or only after doing something specific? Which GPU? Which OS? Come on, you can do better than that... :)

full screen exclusive, 550 ti, Win 7 x64

i didn't test. i just went to full screen like usual and the presentation glitches racked up by 20 or 30 at a time

i also noticed the seek bar appearing is delayed by 2 seconds in DX9 or DX11 path

[aufkrawall]

I think I don't like SuperRes for chroma.
Jinc3 AR:
http://abload.de/thumb/offb1uqw.png (http://abload.de/image.php?img=offb1uqw.png)

Jinc3 AR + SuperRes (default values):
http://abload.de/thumb/onqgugr.png (http://abload.de/image.php?img=onqgugr.png)

Black contoures vanish too much for my taste.

[Sparktank]

SM + any bit + D3D11 + Ordered Dithering
I suppose this is in FSE mode? Or also in windowed mode?

Oops, sorry, yes in FSE.

D3D11 for me requires FSE to function correctly, so I can't really see stats or anything in windowed mode.
In windowed mode + D3D11, I get that short frame repeated glitch.
Looks like 2 frames on infinite loop.

Tried to take screenshot of what's going on, but I suppose I need a fancier screen capture program.
Regular Print Screen doesn't cut it, and MPC-HC screen capture will actually capture the video I am meant to see but without stats.

[madshi]

I can't really find a situation where the new algo beat the old high algo. In some areas of the picture it's a little bit better but on the others areas it's worse.

I used threshold : 0.9 and detail level : 3 to compare and even tested with 1.0 and 2 to see if it makes a difference but I always prefer the high algo. The details are much more well preserved and I don't know why but the high algo just feels more natural to my eyes. With the new one, sometimes I see banding turned into some sort of "clouds".
Thanks, that's good to know.

I agree with your thinking. When I tested your debanding filter, I felt that custom settings needed to stay. But to be honest, I never used it after my testings done...
Same here. Never touched the custom settings again, after we had settled on good presets.

I justed want to add that when 10bit videos and UHD Blu-Ray are available, the debanding filter may have to be fine tuned. But I think we will see that in time :)
You think "low" debanding strength is still too strong for UHD Blu-Ray?

7, Aero enabled. If I choose "disable desktop composition" in MadVR with D3D11 enabled the image just flickers between the first frame and black, while on D3D9 it falls back to the old path and plays normally.
Good catch!! Yes, DWM/Aero off makes D3D11 windowed mode fail, for some reason. Seems to be a limitation of the way I present the frames somehow. That explains some of the issues other users have reported.

I think I don't like SuperRes for chroma.
Jinc3 AR:

Jinc3 AR + SuperRes (default values):

Black contoures vanish too much for my taste.
Well, the question is which image is nearer to the original image, before chroma was subsampled. We can only guess there. It might need some more samples to do a real judgement. Looking at some of my own samples, SuperRes for chroma looks quite promising.

Anyway, you're too fast. At this time I want to concentrate all the feedback on debanding.

Oops, sorry, yes in FSE.

D3D11 for me requires FSE to function correctly, so I can't really see stats or anything in windowed mode.
In windowed mode + D3D11, I get that short frame repeated glitch.
Looks like 2 frames on infinite loop.

Tried to take screenshot of what's going on, but I suppose I need a fancier screen capture program.
Regular Print Screen doesn't cut it, and MPC-HC screen capture will actually capture the video I am meant to see but without stats.
Not necessary. I bet you have DWM/Aero turned off, right? Turn it on, and D3D11 windowed mode will probably work for you.

[Schwartz]

Can you reproduce this? Does it only occur with the XVID? Or also with other files?

I tried reproducing it with a different XVID but couldn't. The new one plays fine. The one that gave me trouble also had some video corruption in it (missing bits and some broken I-Frames, I think?). But it's curious that a wonky XVID could create wonky madVR behaviour isn't it?

[James Freeman]

Bug Report:


Exclusive mode setting -> present several frames in advance,
Has no effect until the video is completely closed (Ctrl-C).

When playback starts with Exclusive mode setting -> present several frames in advance -> Off,
Windowed mode -> PSFIA takes no effect and stays in Old path.

In D3D11 mode:

Changing any setting in madVR settings window causes presentation glitches (you name it) !!!
Worst when I switch D3D11 presentation Off while playing.

FSE causes tons of Presentation Glitches.

Maybe there are more... I'll continue to check.


Old news for some, but I just understood these (I may be mistaken):

Overlay mode is completely separate path than Old or New.
Old/New/Overlay paths are only for D3D9.
No Aero = No New path.

[madshi]

I tried reproducing it with a different XVID but couldn't. The new one plays fine. The one that gave me trouble also had some video corruption in it (missing bits and some broken I-Frames, I think?). But it's curious that a wonky XVID could create wonky madVR behaviour isn't it?
I don't think it was the wonky XVID, but probably something else, but who knows for sure. Wonky files might trouble the decoder, but should not bother madVR.

[Vyral]

Hi madshi !
Thanks for the new version and all those new features.
I will test the 10bit output first because my monitor is supposed to support up to 24bit.
I haven't tested deband yet but I encounter a problem in the settings : we can't create profile group with the new settings.

[Razoola]

@madshi

My fullscreen exclusive mode seek bar issue happens with both DX9 and DX11 paths. I'm using the 32bit build of madVR win7 x64 and MPC-HC 1.7.8. Going back to madvr 087.21 is the only fix I have found so far.

I'm using a gt980 with 350.12 drivers.

[madshi]

I will test the 10bit output first because my monitor is supposed to support up to 24bit.
I haven't tested deband yet but I encounter a problem in the settings : we can't create profile group with the new settings.
Ehm, 24bit usually means 8bit per RGB/YUV component. And 30bit usually means 10bit per RGB/YUV component.

The problem with the profile groups will be fixed in the next build.

My fullscreen exclusive mode seek bar issue happens with both DX9 and DX11 paths. I'm using the 32bit build of madVR win7 x64 and MPC-HC 1.7.8. Going back to madvr 087.21 is the only fix I have found so far.
I think this will be fixed in the next build.

[madshi]

madVR v0.88.2 released

http://madshi.net/madVR.zip

* added D3D11 option "present a frame for every VSync"
* added option to choose medium/high SuperRes error upscaling quality
* pages "image enhancements" and "upscaling refinement" are now profileable
* fixed: dark video in windowed mode when using error dif with 10 bit display
* fixed: FineSharp modes 2 and 3 corrupted image quality
* fixed: some controls (FSE seekbar, and ZoomPlayer controls) didn't work
* fixed: corrupted render times and more when using a lot of rendering steps
* madVR no longer disables desktop composition when D3D11 is activated
* if DWM is disabled, when madVR starts, D3D9 is used instead of D3D11
* if DWM is disabled while madVR renders with D3D11, madVR pauses playback
For those of you with presentation glitches in D3D11 FSE mode, I've added the option "present a frame for every VSync", which makes D3D11 more similar to the "new" D3D9 FSE presentation path. This new options is also enabled by default, just to be safe. Those of you who did *not* have presentation glitches, please uncheck this new option, which should give you slightly more efficient presentation (slightly lower GPU usage), when movie frame rate and display refresh rate don't match.

D3D11 windowed mode seemingly only works with Aero/DWM enabled. So I've made a few adjustments for that. Sorry guys, if you want to use D3D11, it seems you have to enable Aero/DWM. Which is not a problem for Windows 8.1/10, because there it's always enabled, anyway.

[aufkrawall]

My tearing problem is still there, also with new Vsync option.
It btw. doesn't just tear, there is also some kind of jittering.
OSD doesn't count any anomalies.

[madshi]

What does the OSD say? All queues full? Any glitches or frame drops?

[Razoola]

@madshi, can confirm 088.2 fixes my seek bar issue. Now I can finally play with the other new features :)

[Moony349]

All queues are now full with DX11 path; thanks!

No presentation glitches.

[Siso]

D3D9 average rendering time ~ 8,16 ms, present 0.10 ms

D3D11 average rendering time ~ 8.20 ms, present 0.80 ms... rendering queue not filled...

[aufkrawall]

What does the OSD say? All queues full? Any glitches or frame drops?
Everything's fine there. Will try to capture the problem with ShadowPlay.

Edit: I think I forgot this: The problem doesn't seem to occur as long as the window isn't maximized ("borderless fullscreen").

[Razoola]

Quick feedback on dx11 vs dx9

DX11 gave me many presentation glitches regardless of the new setting added in 0882. Its also very jittery, especially after a seek.

I think however I have found a solution for myself that maybe others want to try.

For DX9 I have always used exclusive mode setting;

8
don't flush
don't flush
don't flush
flush & wait (sleep)

That setting causes major issues with DX11 for me. If I change the exclusive mode settings to;

8
flush
flush & wait (sleep)
don't flush
don't flush

Now my DX11 presentation issue fixes itself apart from the odd one here and there after seeking.

[aufkrawall]

ShadowPlay capture showed only tearing and jitter, even if there was none to see while capturing. :angry:
But please see my edit above. My original assumption regarding seeking as the cause was wrong.
It's borderless fullscreen what's causing the behavior, at least with 0.88.2.

[madshi]

Everything's fine there. Will try to capture the problem with ShadowPlay.

Edit: I think I forgot this: The problem doesn't seem to occur as long as the window isn't maximized ("borderless fullscreen").
I see. I've already noticed that DXGI/Direct3D behaves somewhat differently when the window is in fullscreen position, even if FSE is disabled. I think if you right click on the window to open a context menu, the tearing will be gone, right? The problem for me is that there's really not much I can do. I provide DXGI/Direct3D with several frames in advance and tell the API to show each frame for one VSync. All the rest, like VSync control etc is left for DWM to handle. I've no part in that. Basically if you have configured madVR to present 8 frames in advance with a 24Hz display, every frame you see on your display has left madVR's control 333ms earlier. I'm telling you that to stress that it's totally out of my hands.

However, it seems there are only very very few users with the same problem as you. So it must be something specific to your PC somehow, I would guess. You don't have any "adaptive sync" or stuff enabled in your GPU drivers, have you?

[James Freeman]

Thanks!!

This old bug still present:
Exclusive mode setting -> present several frames in advance,
Has no effect until the video is completely closed (Ctrl-C).

When playback starts with Exclusive mode setting -> present several frames in advance -> Off,
Windowed mode -> PSFIA takes no effect and stays in Old path.

In 0.88.2:
FSE does not show that it uses D3D11.


BTW,
10bit works perfectly here with 10bit monitor.
A simple test I use is to play a gray gradient and make it super smooth with High Debanding setting and Disable Dithering;
In 8bit mode I can still see the gradients (just barely) even though they should be super smooth.
In 10bit it is super smooth even without dithering,
Dithered 8bit is as smooth as 10bit; even my i1D3 reads 11bit on dithered 8bit pattern.

So when UHD blu ray comes we definitely have nothing to worry about staying with our old 8bit monitors because dithered 8bit is as good if not better than native 10bit.

[madshi]

Quick feedback on dx11 vs dx9

DX11 gave me many presentation glitches regardless of the new setting added in 0882. Its also very jittery, especially after a seek.

I think however I have found a solution for myself that maybe others want to try.

For DX9 I have always used exclusive mode setting;

8
don't flush
don't flush
don't flush
flush & wait (sleep)

That setting causes major issues with DX11 for me. If I change the exclusive mode settings to;

8
flush
flush & wait (sleep)
don't flush
don't flush

Now my DX11 presentation issue fixes itself apart from the odd one here and there after seeking.
You're near to default settings now. I think the first flush isn't necessary. The default settings are "don't, sleep, don't, don't".

[tobindac]

Panel may be 8 or 10 bit, but picture always processed in digital video processor first. In what bitdepth? Say, 10-16 bits.Thus its best to transport picture in highest bitdepth DVP can accept to get less roundings.

Can anyone confirm that makes sense? Because it seems to me it may end up either adding latency for not reason or even "rounding" to an even more inaccurate representation of the original source.

[JarrettH]

The seek bar is most definitely fixed :thanks:

[madshi]

This old bug still present [...]

In 0.88.2:
FSE does not show that it uses D3D11.
These are not bugs. Some settings need a madVR "recreation" to take effect. That's just the way it is. And FSE mode doesn't show it uses D3D11 because that would make the OSD so big. It shows the bitdepth, though, which D3D9 FSE does not show. So if you see the bitdepth there, it's D3D11.

[aufkrawall]

I see. I've already noticed that DXGI/Direct3D behaves somewhat differently when the window is in fullscreen position, even if FSE is disabled. I think if you right click on the window to open a context menu, the tearing will be gone, right?

Yes, confirmed.


However, it seems there are only very very few users with the same problem as you. So it must be something specific to your PC somehow, I would guess. You don't have any "adaptive sync" or stuff enabled in your GPU drivers, have you?
Well, I'm on Windows 10 beta with WDDM 2.0 beta drivers.
Could be related to bugs or to a DWM change in general, I guess.

Maybe there can be a workaround to keep an invisible context menu opened all the time?
I don't have anything special with Vsync options in the driver, but I have a display connected without scaler and I'm using custom timings.

[mark0077]

The new "present a frame for every VSync" works great for me. D3D11 works perfectly. Heres my average stats after 60 seconds incase its of any use. GTX 980

average stats (D3D9)
rendering 4.75ms
present 0.14ms

average stats (D3D11)
rendering 4.67ms
present 0.10ms

[madshi]

Well, I'm on Windows 10 beta with WDDM 2.0 beta drivers.
Could be related to bugs or to a DWM change in general, I guess.

Maybe there can be a workaround to keep an invisible context menu opened all the time?
I don't have anything special with Vsync options in the driver and have a display connected without scaler and using custom timings.
I rather think it's an OS/GPU bug. Let's wait for final Windows 10 and GPU driver versions before I invest time into this. I don't think an invisible window would work, unfortunately. I don't even have Windows 10 installed yet.

The new "present a frame for every VSync" works great for me. D3D11 works perfectly.
Good to hear. Do you notice any differences between D3D9/D3D11? E.g. is one faster than the other when changing window sizes, moving the media player window, switching between FSE and windowed mode etc?

[Vyral]

Ehm, 24bit usually means 8bit per RGB/YUV component. And 30bit usually means 10bit per RGB/YUV component.


Here is my monitor spec :
Contrast 3 000 : 1 typical
Contrast 12 000 000 : 1 ACR
Brightness 250 cd/m² typical
Display colour 24-bit, 16.7 million

I tested 10bit output and the result wasn't much different from 8bit. Just a bit more colorful.

[tobindac]

16.7mil is by definition the 8bit RGB because (8bit, 8bit, 8bit) -> 16.7mil. 24bit RGB would be an absolute overkill. 10bit is already very hard to justify for regular movie screenings since most people would not see any difference above 6bit (try the madvr settings and set it low, 1bit and 2bit are hilarious, they remind of old video games).

Panel may be 8 or 10 bit, but picture always processed in digital video processor first. In what bitdepth? Say, 10-16 bits.Thus its best to transport picture in highest bitdepth DVP can accept to get less roundings.

Can anyone confirm that makes sense? Because it seems to me it may end up either adding latency for no reason or even "rounding" to an even more inaccurate representation of the original.

[aufkrawall]

I rather think it's an OS/GPU bug. Let's wait for final Windows 10 and GPU driver versions before I invest time into this. I don't think an invisible window would work, unfortunately. I don't even have Windows 10 installed yet.

For the record: It happens also in FSE.

[mark0077]

Good to hear. Do you notice any differences between D3D9/D3D11? E.g. is one faster than the other when changing window sizes, moving the media player window, switching between FSE and windowed mode etc?

I have tested a few things out, results below

Resizing mpc-be window with mouse: No difference between 9 and 11, both seem to stay on a static image (last frame that was shown, being scaled basically) until I stop mouse movement. Not sure should D3D11 immediately / continuously output image while my mouse is moving or maybe this is the players fault? Not a big issue for me either ways, but no difference between the two here.

Moving player window around: No difference both look perfectly smooth to me. No drops or glitches with either.

Switching between FSE and Windowed: I don't notice a difference, both seem very very fast for me, there is that slight flicker where the screen goes black for a few milliseconds, but I guess thats just how it has to be. Speed wise both are equally as fast to my eyes.

If I spot any differences I'll post back.

[James Freeman]

88.2:
Causes 1 glitch each pause in d3d11.

About 10bit (was the last post on previous page):
10bit works perfectly here with 10bit monitor.
A simple test I use is to play a gray gradient and make it super smooth with High Debanding setting and Disable Dithering;
In 8bit mode I can still see the gradients (just barely) even though they should be super smooth.
In 10bit it is super smooth even without dithering because madvR processes in 16bit and outputs in 10bit to the monitor.
Dithered 8bit is as smooth as 10bit; even my i1D3 reads 11bit on dithered 8bit pattern.
So when UHD blu ray comes we definitely have nothing to worry about staying with our old 8bit monitors because dithered 8bit is as good if not better than native 10bit.

madshi,
What is overlay mode? Is it not Direct3D?
Is there any way to go Overlay in window but d3d11 in FSE?

[tobindac]

What's the pipeline to get 10bit output? Does it always require an 10bit source video being played? Or will madvr always represent its output as 10bit?

[Barnahadnagy]

What's the pipeline to get 10bit output? Does it always require an 10bit source video being played? Or will madvr always represent its output as 10bit?

As internal processing is done in floating point, output bitdepth doesn't have much in common with input bitdepth, so it will always output 10bit, if you set it that way.

[Sparktank]

...DWM/Aero off makes D3D11 windowed mode fail...

...I bet you have DWM/Aero turned off, right?...

Sorcery!
This is incredibly new and alien to me!
I've not used Aero since Vista came out.

New data: 0.88.1
Aero enabled, D3D11+SM+Window=success.
No presentation glitches.

(click to view larger)
http://i.imgur.com/kKyNKSMl.png (http://imgur.com/kKyNKSM)

Aero enabled, D3D11+SM+FSE=semi-okay.
Letting it play linear will be fine.
Skipping time will cause presentation glitches.
========================
(previous data for 0.88.1)
========================
madVR v0.88.2 released

...presentation glitches in D3D11 FSE mode, I've added the option "present a frame for every VSync", which makes D3D11 more similar to the "new" D3D9 FSE presentation path.

..Sorry guys, if you want to use D3D11, it seems you have to enable Aero/DWM...

And, now to play with these.
And, then once principle viewing is established, the debanding.

[tobindac]

As internal processing is done in floating point, output bitdepth doesn't have much in common with input bitdepth, so it will always output 10bit, if you set it that way.

Which makes me wonder once more if it would give any advantage to get 10bit output in these cases:

1) 8bit monitor that processes 10bit madvr input from a 10bit video file
2) 10bit monitor that processes 10bit madvr input from an 8bit video file.
3) Any monitor that processes 10bit madvr input from an 8bit video file

Someone mentioned the 1st case may have a benefit because of more accurate rounding but it seems to me it may even create unecessary latencies or add incaccuracies compared to the source.

Intuition says mainly a 10bit monitor should use 10bit output and when it has a 10bit source.

Unless madvr always improves with 10bit output due to its calculations accuracy leading to it.

[Moony349]

Just realized something that, in retrospect, is pretty obvious but just in case anyone else is having high present times:

I had my Nvidia control panel "Max Pre-Rendered Frames" set to 1 to reduce input lag in certain online games.

Changing this to use application setting fixed the present times. This setting had no effect in using d3d9 but apparently does for dx11.

[James Freeman]

Unless you actually see the dithering pattern in 8bit (takes an inhuman superman vision here), dithered 8bit look exactly like 10bit+.
Even a professional calibration meter (i1 Display Pro) reports that it reads 11bit from a dithered 8bit pattern!

[Sparktank]

Aero enabled, D3D11+SM+FSE++time_skip

0.88.2
"present a frame for every VSync"

Confirmed fix working.
Skipping time will no longer cause presentation glitches.

Thanks for the fix!

[baii]

Monitor usually have a build in higher than panel bit depth lut for conversion. You can find Manf advertising 14/16 bit internal lut.

Sent from my 306SH

[Qaq]

Which makes me wonder once more if it would give any advantage to get 10bit output in these cases:

1) 8bit monitor that processes 10bit madvr input from a 10bit video file
2) 10bit monitor that processes 10bit madvr input from an 8bit video file.
3) Any monitor that processes 10bit madvr input from an 8bit video file

Someone mentioned the 1st case may have a benefit
Wrong. Source can be even 4 bit. The trick is madVR converts source YUV > RGB in 16fp. Then it rounds 16fp to either 8int or 10int and adds dither for output. Needless explain in higher output bitdepth we keep more processed data. If display has 16fp processor with perfect dithering, it would be best to output 16fp to it from madVR. Its all about internal conversions and transport, it has nothing to do at all to source and panel bitdepths.